- Kube-controller-manager: Resolves potential issues handling pods with incorrect uids in their ownerReference ([#134658](https://github.com/kubernetes/kubernetes/pull/134658), [@liggitt](https://github.com/liggitt)) [SIG Apps]

        `SerializationContext` argument to enable overriding the default policy
        for the handling external state during iterator checkpointing. This is
        not a backwards compatible change and all subclasses of `IteratorBase`
        *need to be updated* accordingly.
*   `tf.keras`
    *   Add a new `BackupAndRestore` callback for handling distributed training
        failures & restarts. Please take a look at this

## Changelog since v1.16.10

## Changes by Kind

### API Change

- Resolve regression in metadata.managedFields handling in update/patch requests submitted by older API clients ([#91748](https://github.com/kubernetes/kubernetes/pull/91748), [@apelisse](https://github.com/apelisse)) [SIG API Machinery and Testing]

### Feature

- Fixes a race condition in kubelet pod handling ([#94774](https://github.com/kubernetes/kubernetes/pull/94774), [@auxten](https://github.com/auxten)) [SIG Node]

- Fixes error handling in a kubectl method used in downstream packages. ([#108520](https://github.com/kubernetes/kubernetes/pull/108520), [@heybronson](https://github.com/heybronson)) [SIG CLI]

## Dependencies

### Added

- Kube-controller-manager: Fixes a 1.33 regression in daemonset handling of orphaned pods ([#134652](https://github.com/kubernetes/kubernetes/pull/134652), [@liggitt](https://github.com/liggitt)) [SIG Apps]

## Changelog since v1.20.14

## Changes by Kind

### Bug or Regression

- Fixes a panic in kube-scheduler handling pods with invalid selectors ([#107558](https://github.com/kubernetes/kubernetes/pull/107558), [@Zheaoli](https://github.com/Zheaoli)) [SIG Scheduling]
- Fixes a rare race condition handling requests that timeout ([#107461](https://github.com/kubernetes/kubernetes/pull/107461), [@liggitt](https://github.com/liggitt)) [SIG API Machinery]

  /** The concurrency level. */
  final int concurrencyLevel;

  /** Strategy for comparing keys. */
  final Equivalence<Object> keyEquivalence;

  /** Strategy for handling entries and segments in a type-safe and efficient manner. */
  final transient InternalEntryHelper<K, V, E, S> entryHelper;

  /**

- Fixes calculating the requeue time in the cronjob controller, which results in properly handling failed/stuck jobs ([#121327](https://github.com/kubernetes/kubernetes/pull/121327), [@soltysh](https://github.com/soltysh)) [SIG Apps]

variable. The default setting is GODEBUG=cgocheck=1, which implements
reasonably cheap dynamic checks. These checks may be disabled
entirely using GODEBUG=cgocheck=0. Complete checking of pointer
handling, at some cost in run time, is available by setting
GOEXPERIMENT=cgocheck2 at build time.

It is possible to defeat this enforcement by using the unsafe package,